Correlation coefficient via reweighted approach: An application to medical field data

• Main Authors: Najdi, Nur Farah Najeeha, Ahad, Nor Aishah, Okwonu, Friday Zinzendoff, Zulkifli, Malina, Yusof, Zahayu Md
• Format: Conference Proceeding
• Language: English
• Subjects: Correlation coefficients; Datasets; Outliers (statistics); Robustness
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0177227

Pearson correlation coefficient is not robust towards outliers since the component in its standard form comprises the mean, which is susceptible to outliers. Therefore, this study proposed a reweighted approach to robustify the classical correlation coefficient. Performance for both new and classical methods was evaluated by implementing actual data sets from medical field. The data sets applied in this study are categorized into small (n=20), medium (n=40), and large (n=100) groups. From the analysis, the gap between mean and median for each group size is reduced with the implementation of the reweighted approach, which implies that the constraints regarding outliers are overcome. The results revealed that each group indicates positive correlations between variables. The results of classical Pearson correlation are 0.81, 0.68, and 0.74 for small, medium, and large groups. Concurrently, the outcomes for the robust reweighted approach are 0.88, 0.71, and 0.85 for small, medium, and large groups, respectively. The performances of the classical correlation coefficient and robust reweighted approach are approximately equivalent and comparable to each other. As a further matter, the proposed approach has overcome the limitations created by the outliers.